Avalon Water Cooling

[ecliptic]

Are those smooth holes for the water channels with threads on the ends so you can mount the fittings/etc with the threads, or threaded all the way through? 

Looks like 15mm(?) of threads

They are smooth inside, you can see the detail "X" on the drawing. Yes, min 15mm of threads. Americans, be aware that this thread is straight, the fittings need a sealing!

As a conclusion of my design I would say it's over specified but totally does it's job. I am running with 30 degree cooling water but cannot clock my miners higher any ways because that would require a voltage modification. Also the step down power converters are at their limits.
It would totally be possible to produce a cheaper solution, maybe like somebody mentioned before a copper plate with soldered-on pipe. Also the surface smoothness of "Ra 1.6" is way smoother than the original heat sink and makes it expensive.

@Icoin: Your module looks like it only cools the ASIC's and not the voltage converters? Did somebody already build it and use it on an Avalon? I would be curious about other peoples experiences.

I don't feel that there is a market for selling the cooling elements. To much work is involved until you have a running solution, even if you buy the heat sinks. If you want to produce larger quantities a redesign would make sense. Maybe in a way of fixing a PCB on both sides.

The main advantage for me is clearly that the noise is much (really really much!) less and I can keep my rig on the balcony in a hot, humid and dusty environment without worrying too much.

So you would need to both increase the voltage and increase the output current capacity is what you are saying?  I'm not familiar with the Avalon stock design, so I wasn't sure if you could modify anything beyond the clock frequency

[1714155258]

1714155258

[1714155258]

1714155258

[1714155258]

1714155258

[1714155258]

1714155258

[1714155258]

1714155258

[1714155258]

1714155258

[el_rlee]

So you would need to both increase the voltage and increase the output current capacity is what you are saying?  I'm not familiar with the Avalon stock design, so I wasn't sure if you could modify anything beyond the clock frequency

Soldering required for higher rates, and I am even not so sure if it can go much higher... who knows?

[el_rlee]

Any thoughts on running a refrigerant through the blocks?

Condensation. My enemy.

[crumbs]

Any thoughts on running a refrigerant through the blocks?

Condensation. My enemy.

How about a dehumidifier for your balcony & dew (moisture) sensors in a few strategic places?  I guess pretty pointless if you can't clock up for other reasons.

[el_rlee]

Inspired by this thread https://bitcointalk.org/index.php?topic=212090.0 I decided to have a try to improve the resistance between the chips and the aluminium.

The cooling areas of the chips are a bit lower than the board surface and the used thermal conductive foil doesn't show any marks - so it doesn't have real contact.



I removed all my PCB's from the heat sinks once more and applied thermal compound.



That's more like it



I know, a little more wouldn't have hurt probably but I only got 3 syringes and wanted to finish that rig. At least 4g required for one board.

Result: 10GH/s more in --avalon-auto mode (much less HW errors)

[A Meteorite]

Result: 10GH/s more in --avalon-auto mode (much less HW errors)

What clock rate and temps are you hitting with the thermal paste?

[GenTarkin]

avalon switched from a huge sludge of thermal goop(thread u linked) .. ..to the thermal sheet shown in ur picture?

I find it hard to believe that the foil makes little / no contact w/ the chips under the pressure of the screws. It would have to touch them... that thermal foil type stuff is squishy right? ...

Given that, Im sure its just the thermal effeciency working in ur favor of the MX-2 over the horrible effeciency of a thermal pad. =)

Post screenshot of clocks / temps / HW error after several hours of mining(webgui) .. that would be great =)
If its enough of a difference, I may attempt this.

[el_rlee]

I gotta say both my chiller and the pump are at their limit. Water set value is 30 degree as to avoid condensation, but the chiller is any ways running 100% at this hot weather so setting a lower value would have no effect at all.

[GenTarkin]

oh I see, that thermal foil, u tried first on ur WC blocks? I didnt even realize the pictures were a WC block =P
Thats pretty nice, whats the ambient @ that clock?

[el_rlee]

Ambient is sensor one. That thermal pad was delivered with my batch two unit.

[GenTarkin]

Ambient is sensor one. That thermal pad was delivered with my batch two unit.

oh interesting...

So, I was thinking, arent the temp sensors attached to the heatsink and not the asic chips themselves?
Given that...if ur thermal transfer effeciency increased by ur application of MX-2 .. shouldnt the heatsinks be a bit hotter as per the thermal sensors??

[GenTarkin]

An even more optimal application of MX-2 might be to do a thin layer across entire pcb first(kinda how it was stock on early samples) .. then apply a tad to each little thermal pad for the chips.
QFN, iirc is designed to not only release heat through the thermal pad but also to the PCB around it - both work as a cooling element for the chip.

[crazyearner]

Love the effort put into the  water cooling blocks. I would maybe recommend using different cooling fluid than water you have in them tanks and get stuff that is designed to adsorb heat and dispense it. apon hitting cooling as it is actually designed to cool the liquid quicker when travailing into radiator block to cool. Or your next task maybe taking it to a oil cooed rig setup but for water cooling at them terms seems heck of a hell high to me. I am currently working on a few things that I should have up shortly for air cooled for people as I know someone does done the tiniest of mods and now cooling much lower than the water cooling or very close results on air cooled. If I had the money I would go make some water blocks but also have a few ideas too More details later.

[ecliptic]

so the million dollar question is this -- What's the limiting factor here?

Is it just cooling?

Do we need to overvolt?

Is the Avalon designs (and other ones being designed(?)) capable of the increased current draw, not capable of a fast enough transient response, not enough decoupling caps?

Or are we running into a brick wall due to the internals of the ASIC chip itself?


Is there anything stopping a design with water cooling and upgraded power regulator/decoupling from breaking 450 MHz?

[crumbs]

Has anyone decapsulated one of these chips?

[riX]

Has anyone decapsulated one of these chips?

Why would anyone want that? Trying to figure out how hash sha256?

[crumbs]

Has anyone decapsulated one of these chips?

Why would anyone want that? Trying to figure out how hash sha256?

I was simply interested in packaging/heat transfer, that's why i asked in this thread.  I think there was a mention of "hollow" somewhere, a warning against top-side heatsinks, wanted to find out if that was the case.  Also idle curiosity, i like silicon pronz.  

Edit:  Spelingz -> silicone

[GenTarkin]

Has anyone decapsulated one of these chips?

Why would anyone want that? Trying to figure out how hash sha256?

I was simply interested in packaging/heat transfer, that's why i asked in this thread.  I think there was a mention of "hollow" somewhere, a warning against top-side heatsinks, wanted to find out if that was the case.  Also idle curiosity, i like silicone pronz. 

Just wiki or read many pdfs available regarding QFN package design. You will get the general idea, yes under the plastic they are hollow the die sits inside and has "metal support beams" (forgot exact term) that go through the PCB to the thermal pad on the opposite side of the PCB. Heat is distributed to the surface of the PCB & to that metal thermal pad.

[crumbs]

Just wiki or read many pdfs available regarding QFN package design. You will get the general idea, yes under the plastic they are hollow the die sits inside and has "metal support beams" (forgot exact term) that go through the PCB to the thermal pad on the opposite side of the PCB. Heat is distributed to the surface of the PCB & to that metal thermal pad.

I've cracked many open when i was a kid   Never seen one with a cavity inside, that's why i was curious.  I have a hard time imagining why such a cavity would exist, considering how the things are typically made (you can see the resin being poured in one of the frames):


They're all pretty much like this:
So whatever the plastic/resin is used on the topside is in direct contact with the die.  The second gif is, obviously, not to scale, so the plastic on top of the die is not much thicker than the PC board.  In other words, if there is no air cavity (why would there be -- seems much simpler to *not* have one), top-side heatsinks would improve air cooling *a big bunch*

[SolarSilver]

Also idle curiosity, i like silicon pronz.  

Let me get you your daily fix of pr0n:

http://zeptobars.ru/en/read/avalon-bitcoin-mining-unit-rig

No pics of the illustrious cavity though